CN104156623A - Diagnosis method of mechanical composite faults - Google Patents

Abstract

The invention discloses a diagnosis method of mechanical composite faults. The diagnosis method includes the following steps that empirical mode decomposition is used for decomposing a composite fault signal into a plurality of IMF components; the characteristic quantity of each IMF component is extracted, and a fault sample data network model is established; each IMF component is regarded as a community in a network, and similar communities are merged according to the characteristics that internal connection of a community structure of a complicated network is tight and external connection of the community is loose; each community obtained through merging corresponds to one single fault; a merged signal is analyzed, composite fault characteristics are separated into single fault characteristics, and a fault diagnosis structure is obtained. According to the diagnosis method, the composite fault signal is decomposed into the IMF components at different frequencies, each IMF component is regarded as a community structure in the network, and a separated single fault signal is obtained according to merging of the complicated network communities, therefore, composite fault characteristic separation is achieved, and a diagnosis result is obtained.

Description

A kind of diagnostic method of mechanical combined failure

Technical field

The present invention relates to a kind of combined failure diagnostic method, relate in particular to the mechanical combined failure diagnostic method that the separation of combined failure signal characteristic is converted into single failure diagnosis.

Background technology

Along with the progress of industrial technology, large complicated machinery is just towards maximization, complicated development, and whether mechanical movement state well will directly affect commercial production.Yet the equipment part having in engineering reality need to damage to acquire a certain degree and just find and change, at load working condition, extremely under complex situations, during this, may there is various faults the phenomenon of depositing, thereby form combined failure.Due to the equal low-frequency range in analysis frequency of the characteristic frequency of most of combined failure, energy is lower, is often submerged in powerful ground unrest, and the characteristic such as influence each other between fault, is difficult to find out and fault characteristic of correspondence frequency from spectrogram.Therefore, cause difficulty to the comprehensive Accurate Diagnosis of combined failure.

And in recent years, to the development of single fault diagnosis research rapidly, as fft analysis, envelope spectrum analysis, wavelet analysis, spectrum kurtosis etc., these methods all show good effect in single fault diagnosis.Yet, these methods are applied to but can run into many difficulties in combined failure diagnosis, fft analysis and envelope spectrum analysis are when the larger combined failure of the strong and weak gap of tracing trouble, and weak trouble unit is easily submerged among noise, thereby is left in the basket when diagnosis.Wavelet analysis extracts signal characteristic by specific basis function, and in fault diagnosis, widespread use is that single small echo only has a basis function, can only a kind of fault signature of optimum matching, therefore easily attend to one thing and lose sight of another when combined failure feature extraction.Spectrum kurtosis is to choose bandpass filter parameter according to the kurtosis value size of calculating every spectral line, then carry out fault diagnosis, when analyzing combined failure, owing to choosing maximum kurtosis value analysis, cause some fault kurtosis value to be easily missed, thereby be difficult to Accurate Diagnosis, go out each malfunction.Therefore, thereby how combined failure character separation is formed to single failure, carry out fault diagnosis, be difficult point always and lack effective separation method, also become and in fault diagnosis, need the key issue that solves simultaneously.

Summary of the invention

In order to solve the problems of the technologies described above, the invention provides the diagnostic method of the mechanical combined failure that a kind of reliability is high, combined failure character separation is converted into single failure diagnosis, to solve the problem of combined failure difficult diagnosis.

The technical scheme that the present invention solves the problems of the technologies described above comprises the following steps:

(1) application EMD is several IMF components by combined failure signal decomposition;

(2) extract the characteristic quantity of each IMF component, set up fault sample data network model;

(3) each IMF component is considered as to the corporations in network, according to complex network community inside configuration, connects closely, the sparse characteristic of the outside connection of Er Yu corporations, carries out the merging of similar corporations;

(4) merging each corporation of gained are corresponding with single failure;

(5) signal being combined is analyzed, and realizes combined failure character separation and diagnosis.

Technique effect of the present invention is: the present invention's application EMD is several IMF components by combined failure signal decomposition, because the feature of different single failures can be embodied at different frequency range, extract the characteristic quantity of each IMF component, set up fault sample data network model; Then each IMF component is considered as to the corporations in network, according to complex network community inside configuration, connect closely, the sparse characteristic of the outside connection of Er Yu corporations, carry out the merging of similar corporations, merge each corporation of gained corresponding with single failure, the signal being finally combined is analyzed, and realizes combined failure character separation, and obtain diagnostic result, solve combined failure and diagnosed difficult technical matters.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention.

Fig. 2 is combined failure vibration signal and envelope spectrum schematic diagram in the present invention.

Fig. 3 is network corporations cluster coefficients schematic diagram in the present invention.

Fig. 4 is uneven separated with the inner ring combined failure envelope frequency spectrum figure of rotor of the present invention.

Fig. 5 is the rolling bearing inner ring envelope frequency spectrum figure separated with rolling body combined failure in the present invention.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in further detail.

Referring to Fig. 1, Fig. 1 is process flow diagram of the present invention.Specific embodiment of the invention step is as follows:

(1) application EMD (Empirical Mode Decomposition, EMD) is several IMF components by combined failure signal decomposition.EMD method is a kind of brand-new signal Time-Frequency Analysis Method.It is to utilize the variation of signal internal time yardstick to do the parsing of energy and frequency, can be by the intrinsic mode function component that is decomposed into a limited number of linearity, stable state (the Intrinsic Mode Function of the signal adaptive of non-linear, non stationary state, be called for short IMF) IMF1, IMF2, IMF3 ... IMFn, the information of different like this intrinsic modal components reflection different frequency ranges.And combined failure has failure-frequency corresponding to a plurality of fault signatures, each fault frequency range is embodied in different modal components.Utilize this characteristic of EMD, each fault status information in vibration signal can be carried out to separation, the community structure then each IMF component being considered as in complex network carries out corporations' merging.

(2) extract the characteristic quantity of each IMF component, set up fault sample data network model.Pass through field monitoring, obtain combined failure multidate information, combined failure observation signal is carried out to EMD, obtain several IMF components, extract different I MF component fault signature amount, form sample set, each sample abstract be network node, relation between sample and sample is abstract is limit, and the sample of different I MF component just can be abstracted into complex network structures like this, and each IMF component is considered as the corporations in network.

If fault sample collection X={x ₁, x ₂..., x _n, each a sample p attribute, i.e. x _i={ x _i1, x _i2..., x _ip, x _iwith x _jbetween contact similarity a _ij∈ A represents; By each data sample x _ibe considered as network " node ", the contact between data sample is as network " relation ", and data structure can be expressed as fault data network model G (X, A).

Relatively the similarity of different mode can be converted into two vectorial distances of comparison.Generally speaking, a _ijsample x _iand x _jspacing d _ijfunction.The principle of similarity function design is to make network have good block structure (in piece, similarity is approaching as far as possible, and between piece, similarity difference is larger), is defined as:

a _ij＝exp(-0.1*d _ij) (1)

In formula, d _ijfor Euclidean distance tolerance.Obviously, d _ijless, a _ijlarger, show x _iwith x _jbetween similarity larger.Because the similarity of sample self is nonsensical, the similarity that defines self is 0, when i=j, and a _ij=0.Because similarity each other between two nodes equates, i.e. a _ij=a _jiso A is a symmetric matrix.The connection matrix of the network of n node is:

So far, obtain fault data network model G (X, A).

(3) each IMF component is considered as to the corporations in network, according to complex network community inside configuration, connects closely, the sparse characteristic of the outside connection of Er Yu corporations, carries out the merging of similar corporations.Complex network cluster coefficients has disclosed the Clustering features of network, the cluster coefficients C of node i _iin the node that represents to be connected with i, between any two nodes, there is the probability being connected.If node i and k _iindividual node is connected, this k _ibetween individual adjacent node, there is at most k _i(k _i-1) limit/2.Ei is in esse limit number between adjacent node:

$C_i=\frac{2F_i}{k_i(k_i-1)}---\left(4\right)$

The cluster coefficients C of network is the mean value of the cluster coefficients of all nodes in whole network:

$C=\frac{1}{N}\underset{i}{\mathrm{\Σ}}{C}_i---\left(5\right)$

Complex network has community structure characteristic, and when cluster coefficients is larger, the cluster degree of network is higher, is arranged on the different differentiation factors time, calculate each corresponding corporations' cluster coefficients, thereby determine and differentiate the factor work as a _ijbe greater than be set to 1, represent to be related between two nodes; Be less than just be set to 0, represent not contact between two nodes.Network can abstractly be just that the figure being comprised of point set and limit collection represents like this.

(4) merging each corporation of gained are corresponding with single failure.The IMF component that combined failure signal is decomposed through EMD is considered as corporations in network, and carries out the merging of similar corporations.To Z corporations of fault data network model G (X, A) gained, the IMF component number that EMD decomposes.If the X of corporations _p, X _qthe proper subclass of X, i.e. X _p≠ 0, X _q≠ 0, and if there is A _pq=(a _ij), i ∈ X _p, j ∈ X _q, and note so when p ≠ q, for subset X _p, X _qbetween be related to quantity, a _ijduring ∈ A; And during p=q, for the quantity of subset internal relations, a _ij∈ A.Note for the network that has Z community structure, modularity index ^[12-13]q is defined as:

$Q=\underset{p=1}{\overset{c}{\mathrm{\Σ}}}{e}_{\mathrm{pp}}-\underset{p=1}{\overset{c}{\mathrm{\Σ}}}{\left(\underset{q=1}{\overset{c}{\mathrm{\Σ}}}{e}_{\mathrm{pq}}\right)}^2---\left(6\right)$

Wherein, the connection of reflection corporations internal node, embodied the connection of corporations' intermediate node.Obviously it is larger, less, Q is larger, shows that interior nodes density Gao Er corporations of corporations intermediate node connects few.As can be seen here, Q is the quantitative key concept of having portrayed community structure not only, and be widely used in the design of community structure probe algorithm, is a suitable network community structure index.

There is the connection matrix of network G (X, A) of n node suc as formula shown in (3), establish network and formed by Z corporations, can calculate corresponding e _ij∈ E ⁿ, note

If the p of corporations and q are merged, will cause that modularity index Q is changed to

$\stackrel{\‾}{Q}=\underset{m=1}{\overset{z}{\mathrm{\Σ}}}{e}_{\mathrm{mm}}+e_{\mathrm{pq}}+e_{\mathrm{qp}}-\underset{m=1}{\overset{z}{\mathrm{\Σ}}}{\left(\underset{i=1}{\overset{z}{\mathrm{\Σ}}}{e}_{\mathrm{mi}}\right)}^2-2\underset{i=1}{\overset{z}{\mathrm{\Σ}}}{e}_{\mathrm{pi}}=Q+2e_{\mathrm{pq}}-2\underset{i=1}{\overset{z}{\mathrm{\Σ}}}{e}_{\mathrm{qi}}\underset{i=1}{\overset{z}{\mathrm{\Σ}}}{e}_{\mathrm{pi}}---\left(8\right)$

Here note:

Wherein

In formula, e _pq=e _qpfor subset X _pand X _qbetween similarity; e _pifor subset X _pand X _ibetween similarity; e _qifor subset X _qand X _ibetween similarity.Visible, merge subset X _qand X _pcan make the value of Q occur variation.Consider that the larger Clustering Effect of Q value is better, as long as meet just explanation is by X _qand X _pmerging is effectively, further it is optimum and effective just guaranteeing this time to merge.Any Liang Ge corporations are merged and all can calculate corresponding modularity and merge index and change be that each union operation can calculate before:

Merge and meet subset X _qand X _p.If have a plurality of be greater than zero only need to merge two maximum subsets, if illustrate that merging process is own through finishing.

Signal after separation is carried out to envelope spectrum analysis, corresponding with single failure signal.

(5) signal being compared to analysis, is single failure feature by combined failure character separation, obtains fault diagnosis result.

Content in conjunction with the inventive method provides following embodiment:

The present invention chooses imbalance-rolling bearing inner ring, two kinds of combined failures of rolling bearing inner ring-rolling body as embodiment, and bearing designation is ER10K, increases by two 6 grams of small screws on rotating disk, and structure rotor unbalance, gathers combined failure vibration signal.Performing step is as follows:

Step 1: on mechanical fault integrated simulation experiment bench, normal bearing is installed in one end, axle two ends, the other end is installed inner ring fault bearing; The rotating speed of axle is 2396r/min, and sample frequency is 16384Hz, according to fault characteristic frequency computing formula, and rolling bearing inner ring fault characteristic frequency f under this operating mode as calculated _i=197.1Hz, the characteristic frequency of rotor unbalance fault is 40Hz, gather rotor unbalance-rolling bearing inner ring combined failure vibration signal as figure (2), and combined failure is carried out to EMD decomposition, extract and decompose each IMF component characteristic quantity later, by formula (1), set up fault sample network model;

Step 2: the network clustering coefficient in calculating fault sample network model is as figure (3), thereby definite differentiation factor when time most of corporations cluster coefficients values decline more obviously, because cluster coefficients is higher, the cluster degree of network is better, therefore, this is selected herein thereby the fault data network model of setting up, its network connection matrix A.

Step 3: redundancy corporations merge, and each IMF component is considered as to community structures different in fault sample network, calculates corporations between two by formula (9) and merges index variation, change indicator after calculating merges obtain formula (10), merge and meet two IMF components, if having a plurality of be greater than zero only need to merge two maximum IMF components, if all be less than 0 explanation and merge oneself end.As space is limited, only having listed index after last merging changes.

By above-mentioned amalgamation result, can be found out, IMF1, IMF2, IMF3, IMF4, IMF5 combines, and remaining IMF component combines.

Step 4: as shown in Figure 4, for the signal after separation, carry out envelope spectrum analysis, as shown in Fig. 4 (a), can see spectrum peak 192.1Hz, this value approaches very much with inner ring fault characteristic frequency 197.1Hz and 2 frequencys multiplication, 3 frequencys multiplication are high-visible, shows to exist in bearing inner ring fault.As Fig. 4 (b), spectrum peak 40Hz and frequency multiplication are fairly obvious, show that rotor exists imbalance fault.This conforms to the failure condition that experiment arranges.

Step 5: for the validity of further verification method, structure rolling bearing inner ring and rolling body combined failure, the rotating speed of axle is 2095r/min, and sample frequency is 16384Hz, and under this operating mode, rolling bearing inner ring fault and rolling body fault characteristic frequency are respectively f as calculated _i=173.2Hz, f ₀=69.7Hz, carries out envelope spectrum analysis as shown in figure (5) for the signal after separation.By Fig. 5 (a), can be found out, spectrum peak 160.1Hz and inner ring fault characteristic frequency 150.2Hz approach, and 2 frequencys multiplication and 3 frequencys multiplication (320.2Hz ≈ 2f _i, 480.2Hz ≈ 3f _i), show to exist in bearing inner ring fault.In Fig. 5 (b), spectrum peak 64.03Hz and rolling body fault characteristic frequency 60.5Hz are very approaching, and have 2 fairly obvious frequencys multiplication and 3 frequencys multiplication (128.1Hz ≈ 2f _i, 192.1Hz ≈ 3f _i), this result shows to have rolling body fault in bearing, and this conforms to the failure condition that experiment arranges.

Claims (4)

1. a diagnostic method for mechanical combined failure, comprises the following steps:

(1) application EMD is several IMF components by combined failure signal decomposition;

(2) extract the characteristic quantity of each IMF component, set up fault sample data network model;

(3) each IMF component is considered as to the corporations in network, according to complex network community inside configuration, connects closely, the sparse characteristic of the outside connection of Er Yu corporations, carries out the merging of similar corporations;

(4) merging each corporation of gained are corresponding with single failure;

(5) signal being combined is analyzed, and by combined failure character separation, is single failure feature, obtains fault diagnosis result.

2. the diagnostic method of mechanical combined failure according to claim 1, described fault sample data network modeling procedure is:

If fault sample collection X={x ₁, x ₂..., x _n, each a sample p attribute, i.e. x _i={ x _i1, x _i2..., x _ip, x _iwith x _jbetween contact similarity a _ij∈ A represents;

By each data sample x _ibe considered as network " node ", the contact between data sample is as network " relation ", and data structure can be expressed as the undirected network G of weighting (X, A);

Relatively the similarity of different mode can be converted into two vectorial distances of comparison, generally speaking, and a _ijsample x _iand x _jspacing d _ijfunction, the principle of similarity function design is to make network have good block structure, is defined as:

a _ij＝exp(-0.1*d _ij)

In formula, d _ijfor Euclidean distance tolerance, d _ijless, a _ijlarger, show x _iwith x _jbetween similarity larger; The similarity of self is 0, when i=j, and a _ij=0, because similarity each other between two nodes equates, i.e. a _ij=a _jiso A is a symmetric matrix, the connection matrix of the network of n node is:

So far, obtain fault data network model G (X, A).

3. the diagnostic method of mechanical combined failure according to claim 1, the IMF component that described combined failure signal decomposes through EMD is considered as the corporations in network, carry out the merging of redundancy corporations, according to the corresponding modularity of any Liang Ge corporations' joint account, merge index and change be that each union operation can calculate before:

Until all all be less than till zero, redundancy corporations merge end.

4. the diagnostic method of mechanical combined failure according to claim 3, it is that the IMF component of merging is combined into the analysis of row envelope spectrum again that described redundancy corporations merge, and will not have separative combined failure signal to carry out envelope spectrum analysis and compares.